Digital screen exposure has become ubiquitous, yet its causal impact on self-regulation traits remains poorly understood due to methodological limitations in observational research. This study develops a methodological innovation by integrating Mendelian randomization (MR) with advanced multivariable modeling and machine-learning instrument selection. Unlike conventional MR applications primarily in biomedical research, our framework extends genetically informed causal inference into computational social systems. Specifically, we propose a multivariable MR approach enhanced by least absolute shrinkage and selection operator (LASSO)-based instrumental variable selection, which disentangles the independent causal effects of multiple, genetically correlated digital exposures. Using genome-wide association summary data from over 450 000 individuals from the UK Biobank, we investigate the behavioral effects of mobile phone use, computer use, computer gaming, and television viewing on self-regulation traits. The results show that mobile phone and computer use increase risk-taking tendencies, television viewing is associated with lower risk-taking scores and higher perseverance scores, possibly reflecting the passive and sequential nature of this activity, while computer use and gaming are linked to diminished endurance. These findings not only uncover causal pathways between digital activity patterns and psychological traits but also demonstrate a robust methodological advance for applying MR in complex behavioral systems. By demonstrating how screen exposure causally shapes core personality traits, this study provides actionable insights for technology designers to develop personality-aware digital interfaces, for policymakers to establish evidence-based screen time guidelines differentiated by media type, and for parents and educators to implement personalized digital engagement strategies that account for individual differences in self-regulation capacity.

In the digital era, deepfakes have become a critical threat to social media platforms and online information systems, undermining public trust and enabling misinformation campaigns at unprecedented scales. The advancements of artificial intelligence have promoted the rapid development of deepfake technology which enables the manipulation of visual content in highly realistic ways. While this technology holds potential for creative applications, it poses serious risks to digital platforms, public trust, and authenticity to online information. This growing concern demands efficient deepfake detection mechanisms. However, many studies primarily focus on facial manipulations, overlooking nonfacial alterations thereby leaving critical vulnerabilities unaddressed. So, this article proposes a novel approach that combines custom autoencoder (CA) based watermarking with postquantum cryptographic technique to detect the facial and nonfacial deepfakes. The proposed system detects deepfakes by extracting facial region as region of interest (ROI) and nonfacial region as nonregion of interest (NROI) using RetinaFace. A <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$\text{SHA}_{256}$</tex-math> </inline-formula> hash, h<inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">${}_{1}$</tex-math> </inline-formula> is generated for NROI and encrypting it using <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">N</i>th-degree truncated polynomial ring (NTRU). This encrypted h<inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">${}_{1}$</tex-math> </inline-formula> is embedded into ROI using CA to generate watermarked image. During extraction, NROI of watermarked image is reprocessed to calculate a new hash, h<inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">${}_{2}$</tex-math> </inline-formula>. The encrypted h<inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">${}_{1}$</tex-math> </inline-formula> is extracted from ROI and decrypted to retrieve h<inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">${}_{1}$</tex-math> </inline-formula>, which is then compared with h<inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">${}_{2}$</tex-math> </inline-formula>. Any mismatch reveals potential deepfakes. Experimental validation shows that our method achieves a high peak signal-to-noise ratio of 80.29 dB and an structural similarity index measure of 0.90, indicating that the watermark is invisible. It achieves 96% detection accuracy on our test set. Unlike most current methods that focus only on the face, our approach also identifies tampering with the body or background. Additionally, by using the NTRU encryption algorithm, it is protected against future quantum computer attacks. This makes it a practical and secure solution for safeguarding real-world digital media.

In the era of digital intelligence, wearable devices act as pivotal tools for monitoring mental health data. These data facilitate the delivery of targeted support via mental health intervention platforms. As this integrated model gains traction, the selection of appropriate platforms has become essential to ensure effective mental health services for users. Given the involvement of large-scale users with conflicting opinions, this selection issue constitutes a large-scale group decision-making problem, underscoring the need for group consensus. Within this context, effectively addressing trust semantics and minority opinions emerges as two major challenges. To address these challenges, this article explores the cooperative mental health monitoring group consensus based on dynamic trust semantics and behavioral cultural modeling. First, to address trust propagation and evolution within dynamic trust semantics, an ordered trust propagation method and a trust evolution model are constructed to establish a complete and reliable social network among users. Second, a comprehensive index is designed based on opinion similarities and trust relationships to guide the Leiden community detection, thereby reducing the user dimensionality. Third, an objective minority opinions handling method is explored. Specifically, the user weight identified as having minority opinions is increased, while an adjustment strategy that considers the confidence level is applied to remaining users. Finally, the effectiveness and robustness of the proposed group consensus method for mental health monitoring are demonstrated via extensive experimental validation.

Multimodal recommendation is a crucial technology on social media platforms. It is widely applied in scenarios such as product recommendation and advertising delivery. However, existing multimodal recommendation approaches often overlook invariant semantic features that persist across modalities, leading to decreased robustness and generalization. To address this limitation, we propose <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">GILMRec</i>, a novel <underline xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">g</u>raph <underline xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">i</u>nvariant <underline xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">l</u>earning-based <underline xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">m</u>ultimodal social media <underline xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">rec</u>ommendation framework. The GILMRec introduces an invariant feature learning strategy to extract invariant features separately from visual and textual modalities and employs an attention-based fusion mechanism to integrate them into a unified embedding. Specifically, we construct modality-specific similarity graphs and apply top-<inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$t$</tex-math> </inline-formula> neighbor aggregation, enhancing the consistency of invariant features while effectively suppressing modality-specific noise. Extensive experiments on three Amazon benchmark datasets and a large-scale dataset [baby, sports, clothing, and compact discs (CDs)] demonstrate that GILMRec consistently outperforms twelve state-of-the-art baselines. The results confirm the efficiency of invariant features in capturing robust multimodal representations and improving recommendation performance, particularly in sparse data scenarios.

Online social networks (OSNs) have become increasingly relevant for studying social behavior and information diffusion. Nevertheless, they are limited by restricted access to real OSN data due to privacy, legal, and platform-related constraints. In response, synthetic social networks serve as a viable approach to support controlled experimentation, but current generators reproduce only topology and overlook attribute-driven homophily and semantic realism. This work proposes a homophily based algorithm that produces synthetic microblogging social networks such as <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$\mathbb{X}$</tex-math></inline-formula>. The model creates a social graph for a given number of users, integrating semantic affinity among user attributes, stochastic variation in link formation, triadic closure to foster clustering, and long-range connections to ensure global reachability. A systematic grid search is used to calibrate five hyperparameters (affinity strength, noise, closure probability, distant link probability, and candidate pool size) for reaching five structural values observed in real social networks (density, clustering coefficient, large connected component proportion, normalized shortest path, and modularity). The framework is validated by generating synthetic OSNs at four scales (<inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$10^{3}$</tex-math></inline-formula>–<inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$10^{6}$</tex-math></inline-formula> nodes), and benchmarking them against a real-world Bluesky network comprising four million users. Comparative results show that the framework reliably reproduces the structural properties of the real network. Overall, the framework outperforms leading importance-sampling techniques and classical zero-shot generators applied to the same baseline. The generated graphs capture topological realism and yield attribute-driven communities that align with sociological expectations, providing a realistic, scalable testbed that liberates social researchers from relying on live digital platforms.

Detecting multimodal deepfakes has become a pressing concern due to the rising sophistication of generative techniques capable of creating highly convincing visual-speech synchronized deepfakes. As these manipulated contents proliferate on social media, remote education, smart meetings, smart homes, and virtual reality (VR), their increasing realism significantly impacts individuals, underscoring the urgent need for effective detection methods to curb their deceptive spread. Several research efforts have aimed to detect visual–audio deepfakes by examining inter-modal correlation and inter-class consistency. However, optimal detection results have yet to be achieved. Studies have indicated that detectors learn forged characteristics specific to particular generative methods and hide the real class in the visual modality, leading to a bias towards forgery distributions and degrading detection performance. Since audio modality data is a 2-D signal, we consider this bias in the audio modality to be more serious than in the visual modality, a conclusion we have verified through experiments. Therefore, an audio encoder that learns a universal audio distribution is indispensable for effective audio modality deepfake detection. In this work, we propose a new visual–audio deepfake detection method visual–audio multitask transformer with specialty audio encoder (VATS), designed to learn a more robust and comprehensive representation, thereby mitigating overfitting. The structure includes unimodal encoders for preliminary encoding and a dual-stream structure based on a multimodal framework for multilevel interaction and precise detection. To prevent the audio encoder from focusing excessively on specific identities or generator textures, we employ a pretrained speech encoder in the audio stream, while utilizing a linear projection encoder in the visual stream to focus on the fine-grained features of each video frame rather than the semantic information of the graphic. Our method comprises three key components: the fast temporal multimodal alignment and embedding module (FTAE), the modality interaction block (MIB), and the interpretable manipulated location module (IML). Experimental results on FakeAVCeleb demonstrate that our proposed method outperforms previous works in multimodal deepfake detection.

In modern mobile edge computing (MEC) systems, effectively integrating service caching, task offloading, and service delivery is crucial for ensuring high network quality and enhanced user experience. In this article, we introduce a novel joint optimization framework that employs Uncrewed aerial vehicles (UAVs) as dynamic edge computing nodes, collaborating with service collection units (SCUs) that aggregate service requests from Internet of Things (IoT) devices. In the proposed system, UAVs equipped with small-capacity edge servers can process computational tasks locally or forward them to a base station (BS) via the backhaul when necessary. Both user layers (UEs) and UAVs dynamically cache programs using an adaptive algorithm based on task frequency and program weight, which maximizes cache hit rates and expedites service execution. A joint optimization problem is formulated to minimize the overall end-to-end latency through coordinated task offloading, CPU allocation, caching, and UAV scheduling, subject to energy, storage, and mobility constraints. Our resource-aware task offloading and UAV trajectory planning further enable efficient service responsiveness and minimize reliance on the backhaul link. The overall objective is to reduce maximum task latency and backhaul congestion across the network. Extensive simulation results show that our framework significantly outperforms standard baselines, including random caching, most caching, and calculation delay and energy consumption multiobjective optimization problem (CDECMOP), achieving up to <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$18.5\%$</tex-math> </inline-formula> lower energy consumption and <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$23.6\%$</tex-math> </inline-formula> lower latency across diverse network scenarios. These findings underscore the effectiveness of the proposed UE-UAV cooperative caching and computation strategy for scalable, energy-efficient, and low-latency UAV-assisted MEC systems.

In contemporary society, online public opinion exerts significant influence, and guiding its trajectory is essential for maintaining social stability. To capture the interaction and evolution of competing guiding information during topic dissemination, a competitive guidance topic propagation model based on evolutionary game theory is developed. First, to address the interactive influence of multiple topic features, a multifactor analysis framework is established. Based on positive and negative sentiment, a metric for topic controversy is designed. Considering the effectiveness of multiple regression in measuring diverse features, multiple regression is introduced to evaluate the driving forces of different topic characteristics. This approach allows a detailed analysis of how various features affect both the dissemination impact and the dissemination potential. Second, to address the competition and cooperation among different types of guidance information, evolutionary game theory is introduced to design a user participation-driven game mechanism. The mechanism is used to analyze the competitive and cooperative dynamics between diverse guidance messages. It reveals the strategic landscape of positive and negative information dissemination and provides a theoretical foundation for modeling user state transitions. Finally, to address the dynamic nature of user state changes and considering the similarity between infectious disease transmission and topic propagation, a suspicion–hesitation–positive and negative information–immunity model is proposed. The model allows fine-grained modeling of user state transitions and provides effective theoretical support for public opinion management. Experimental results demonstrate that the proposed model effectively fits the propagation trends of guiding topics on the given dataset and captures competitive and cooperative relationships among different types of guiding information. Compared with baseline models, performance improves by at least 2.8%, providing a solution for modeling and analyzing the propagation process of guiding topics.

Depression has become one of the most serious mental illnesses, leading to a substantial decline in quality of life, an elevated risk of suicide, and significant societal challenges. Despite significant progress in the application of deep learning for depression diagnosis, most prevalent methods rely on correlative rather than causal features, limiting their accuracy and interpretability. Here, we propose a causal graph learning (CGL) method for the hierarchical diagnosis of depression. Specifically, we first construct a novel depression facial graph (DFGraph) structure based on a prior knowledge, which collects information about subjects’ facial cues. Our CGL model leverages the DFGraph structure and incorporates a built-in masking mechanism, which is designed to effectively differentiate causal features from confounding ones. It employs backdoor adjustment techniques, which control for confounding variables by blocking noncausal paths, to identify and select pertinent causal features, thereby enhancing the accuracy of the hierarchical diagnosis of depression. We conducted extensive experiments on the collected depression dataset. Our results show that the proposed method provides better results and interpretability is further improved compared to the publicly available baseline.